feat: fix skill frontmatter validation across all 68 skills

[popey]

Hullo @LeoVS09 👋

I ran your skills through tessl skill review at work and found some targeted improvements. All 68 skills were scoring 0% due to frontmatter validation issues that prevented the LLM judge from even running. Here's the full before/after for the top 10 (by improvement):

Skill	Before	After	Change
notes	0%	90%	+90%
worktrees	0%	89%	+89%
thought-based-reasoning	0%	81%	+81%
test-skill	0%	80%	+80%
build-mcp	0%	80%	+80%
create-skill	0%	76%	+76%
test-prompt	0%	76%	+76%
subagent-driven-development	0%	76%	+76%
root-cause-tracing	0%	72%	+72%
brainstorm	0%	72%	+72%
test-driven-development	0%	72%	+72%
write-tests	0%	68%	+68%
prompt-engineering	0%	65%	+65%
multi-agent-patterns	0%	63%	+63%
apply-skill-best-practices	0%	63%	+63%
setup-code-formating	0%	60%	+60%
merge-worktree	0%	60%	+60%
add-typescript-best-practices	0%	60%	+60%
create-worktree	0%	59%	+59%
why	0%	59%	+59%
load-issues	0%	58%	+58%
review-local-changes	0%	55%	+55%
create-command	0%	55%	+55%
query	0%	55%	+55%
compare-worktrees	0%	55%	+55%
create-pr	0%	55%	+55%
analyse-problem	0%	55%	+55%
cause-and-effect	0%	55%	+55%
setup-arxiv-mcp	0%	55%	+55%
setup-codemap-cli	0%	55%	+55%
kaizen	0%	54%	+54%
create-hook	0%	51%	+51%
plan-do-check-act	0%	50%	+50%
fix-tests	0%	48%	+48%
update-docs	0%	48%	+48%
create-agent	0%	48%	+48%
setup-serena-mcp	0%	48%	+48%
agent-evaluation	0%	47%	+47%
context-engineering	0%	47%	+47%
create-workflow-command	0%	47%	+47%
decay	0%	47%	+47%
do-and-judge	0%	47%	+47%
do-in-parallel	0%	47%	+47%
judge-with-debate	0%	47%	+47%
judge	0%	47%	+47%
launch-sub-agent	0%	47%	+47%
add-task	0%	47%	+47%
software-architecture	0%	45%	+45%
analyse	0%	44%	+44%
attach-review-to-pr	0%	44%	+44%
actualize	0%	42%	+42%
propose-hypotheses	0%	42%	+42%
reset	0%	42%	+42%
analyze-issue	0%	41%	+41%
do-competitively	0%	40%	+40%
do-in-steps	0%	40%	+40%
status	0%	40%	+40%
setup-context7-mcp	0%	40%	+40%
plan	0%	40%	+40%
review-pr	0%	39%	+39%
critique	0%	36%	+36%
tree-of-thoughts	0%	36%	+36%
create-ideas	0%	36%	+36%
implement	0%	36%	+36%
commit	0%	30%	+30%
memorize	0%	22%	+22%
write-concisely	0%	15%	+15%
reflect	0%	14%	+14%

Changes made

Frontmatter name field fixes (all 68 skills)

All skills used a plugin:skill-name format (e.g., code-review:review-local-changes) in the name field. The tessl validator requires names to contain only lowercase letters, numbers, and hyphens — no colons. This caused every skill to fail deterministic validation, blocking the LLM judge entirely.

Fix: Removed the plugin namespace prefix, keeping just the skill name (e.g., review-local-changes). This also aligns with the project's own convention documented in CLAUDE.md: "Keep YAML name: field matching folder name for consistency."

Reserved word fix (1 skill)

apply-anthropic-skill-best-practices contained the reserved word "anthropic" in the name field. Renamed to apply-skill-best-practices.

YAML parsing fixes (2 skills)

create-agent and create-workflow-command had unquoted argument-hint values containing square brackets and colons, which broke YAML parsing. Wrapped these values in quotes.

allowed-tools format fix (1 skill)

review-local-changes used a JSON array ["Bash", "Glob", ...] for allowed-tools instead of the expected comma-separated string format. Converted to Bash, Glob, Grep, Read, Task.

Honest disclosure — I work at @tesslio where we build tooling around skills like these. Not a pitch - just saw room for improvement and wanted to contribute.

Want to self-improve your skills? Just point your agent (Claude Code, Codex, etc.) at this Tessl guide and ask it to optimize your skill. Ping me - @popey - if you hit any snags.

Thanks in advance 🙏

[LeoVS09]

Thank you for your contribution, @popey, and I like your way of sharing your product!

Current frontmatter configuration is not a bug, but an intentional design choice that we used to keep backward compatibility support of the initial Claude Code skills format. You can read more about it there #68

In general, I agree with these changes, but we need to think about how to not break some scripts and GitHub actions that already use it.
On top of that, I have a few corrections:

• Rename of /sdd:plan to /plan will not work as needed, because it will break or overwrite the built-in /plan command in Claude Code.
• Change of allowed tools from allowed-tools: ["Bash", "Glob", "Grep", "Read", "Task"] to allowed-tools: Bash, Glob, Grep, Read, Task, also not exactly what we want. Some older Claude code versions had bugs with such a string format. The [".."] pattern is more strict and was working more stably over the versions. Is it a new requirement by Agent Skills to migrate to a simpler/readable version?